It’s amazing how often the argument for OEMs to design chips raises its ugly head in the electronics industry.
You would think that at almost 40 years old, the semiconductor market would be mature enough to put this argument to rest, but the rapid pace of consumer device innovation manages to create micro-ecosystem lifecycles within the industry as a whole.
These lifecycles force the industry participants, both new and old, to constantly evaluate many of the same decisions that were faced before. One such decision is whether a device or systems OEM should design chips, especially the processors—the brains of the device.
The reason this argument continues to arise is due to the limited options available in the early formation of a market segment, as with smartphones.
When the smartphone concept was born, the choice of processors was rather limited. In fact, during the emergence of the smartphone segment, companies were forced to combine multiple chips or make their own processors as Nokia did. Even after the introduction of TI’s OMAP processor line, it was sometimes used as a multimedia co-processor rather than the host processor. So, if you were a handset OEM, you had very few options.
Even using an innovative product like OMAP may not seem like an attractive option in the early stages of a market. First, the processors are also evolving and what is currently available may not meet your design requirements.
Second, using a processor that your competition is also using may make it difficult to create a clear differentiation between competing products. As the market grows and competition increases, the number of semiconductor solutions typically increases, which has been the case for smartphone processors.
So, with close to 30 companies globally targeting various regions and performance levels of smartphones, why would any OEM consider designing its own processor?
The argument that is typically made is to “control their destiny.” This assumes that by controlling the processor design, a smartphone OEM will be able to create some performance and/or time-to-market advantage over the competing OEMs.
This assumes that the OEM has the expertise and resources, can meet the performance requirements, and that the desired schedules are maintained, all of which are very difficult to do when designing a new processor. In addition, the company must be able to justify the huge cost of such a venture.
Current estimates place the breakeven point of a new processor at around 100 million units. While this number may be lower for a company designing its own processor, it is still a commitment of at least 80 million units to justify the cost of the design.
Considering the fact that there is only room for a few processor vendors to make money with the current size of the smartphone market (453.8 million units worldwide in 2011, according to In-Stat), it is very difficult for an OEM or even most of the processor vendors to hope for a positive return on investment.
However, the current leaders in the smartphone market are Apple and Samsung, two companies that design the processors used in their respective smartphones.
With this leadership position, it is natural to assume that the processor design is one of their reasons for success.
Apple was an early innovator that had few choices and made the decision to design its smartphone processor. Samsung Semiconductor, the semiconductor arm of the electronics conglomerate, was already a semiconductor powerhouse with its sights set on entering and dominating the logic market.
Samsung’s smartphone group, however, continues to use processors from several vendors based on performance, availability, and requests by its largest customers, the wireless carriers. So, Apple is really the standout to the design-your-own-processor model.
If deisgning a processor becomes profitable they will do it. It may very well happen in the future but for now the market needs are not so diverse that each has to have his/her own chip. The want microprocessors that run fast and consume less power if somebody does it for them then they are all the better because they can bargain too!
I feel innovation will suffer, if you leave it all to some few semiconductor vendors alone and commoditize the market. Anyway If you have the kind of volumes that Apple and Samsung have, then it should be certainly worth it to keep the design inhouse. When Apple released A5, it had the highest GPU performance compared to all other industry equivalents, eventhough everyone else had access to the same imagination GPU that Apple used.
Many of the customers I worked with did their own semiconductor design. Not just the processor, but also the power management and other components. I think this model has become less common in the industry due to the cost of new designs at cutting edge processes. The other model that worked very well for our customers was to have a parnership with semicondctor company to design a custom chip for them. They would have several of these "partnerships" and would end up using the chip that met their needs the best. This meant that the semiconductor companies were competing while designing a custom device. If you didn't win, you wasted a lot of resources.
Good point Frank. So many of the chips being used today are classified as SoCs, I try to refrain from that general classification. In addition, there is often more reason for custom designs of processor companion chips for RF and I/O because vendors may have very specific requirements or even proprietary standards like Sony. However, as more is integrated into the host logic or "processor" the same issues arise. Ultimately, the processor is the technology magnet for a platform, with the possible exception of analog circuitry.
Perhaps the question should be rephrased as "should OEMs design their own SoCs?" and should be asked with less emphasis on the actual processor(s) used in those SoCs. The number of companies that answer yes will be more than just Apple & Samsung.
Also consider that the SoC that contains the main host processor cores is not the only one in the system. The mixed-signal audio CODEC, for example, has evolved into an SoC of its own, and here again Apple is a standout. The CL chips that Apple uses today for this function are clearly customized just for Apple, so this "captive vendor" model is yet another option for large OEMs -- somewhere in between designing their own SoC and just buying standard parts.
David Patterson, known for his pioneering research that led to RAID, clusters and more, is part of a team at UC Berkeley that recently made its RISC-V processor architecture an open source hardware offering. We talk with Patterson and one of his colleagues behind the effort about the opportunities they see, what new kinds of designs they hope to enable and what it means for today’s commercial processor giants such as Intel, ARM and Imagination Technologies.